GigaOM » exascale

Huawei finds favor at CERN: researchers sign up for more UDS cloud storage

David Meyer — Fri, 01 Feb 2013 11:54:05 +0000

China’s Huawei may find business tough in the U.S. due to suspicions over its motives, but its cloud efforts are clearly appreciated elsewhere. A year after it started working with CERN on cloud storage – something of a priority for a research organization that generates more than 25 petabytes of physics data each year – Huawei has become an official CERN openlab partner, with at least three more years’ collaboration now assured.

The new arrangement was announced on Thursday, along with confirmation of Russia’s Yandex becoming an openlab associate in the field of data processing. Huawei’s involvement is a bigger deal than that, as it puts the Chinese firm on a par with Intel, HP, Oracle and Siemens, all of which work particularly closely with CERN to see how their technologies can help with the Large Hadron Collider experiments.

In Huawei’s case, the company is contributing its self-healing UDS cloud storage system for use and validation. UDS is targeting the upcoming exascale (an exabyte is roughly a million terabytes) era with a mass object-based storage infrastructure that uses ARM’s energy-efficient processor architecture alongside cheap SATA disks. It also offers Amazon S3 API compatibility and claims eleven-nines (99.999999999 percent) reliability, so users theoretically don’t need to back up data stored in a UDS-toting cloud.

UDS provides a bit of insight into how openlab works. Huawei first delivered a 384-node version of UDS to CERN in early 2012, after which the researchers played around with it for three months. In September of that year, Huawei released UDS to the general enterprise market (in more normal eight-node configurations). The benefits for both sides of this partnership are clear: CERN has to push technological limits in order to handle the very big data generated by the LHC, and Huawei gets both valuable feedback from the researchers and a glowing report card to show off to the wider world.

As for the next steps in this partnership, CERN has now hired two computer scientists to work with Huawei on its implementation there, and more UDS storage systems will be deployed at the Swiss facility in the next few months.

Related research and analysis from GigaOM Pro:
Subscriber content. Sign up for a free trial.

Systems to handle big data might be this generation’s moon landing

Stacey Higginbotham — Mon, 02 Apr 2012 04:00:47 +0000

An effort to build a radio telescope that can see back 13 billion years to the creation of the universe is prompting a five-year €32 million ($42.7 million) effort to create a low-power supercomputer and networks to handle the data the new radio telescope will generate. The DOME project, named for a mountain in Switzerland and the covering of a telescope, is the joint effort between IBM and the Dutch space agency ASTRON to build such a network and computer.

There are three problems with building a telescope capable of reading radio waves from that far out in deep space (actually there’s a real estate problem too, because the array will require millions antennas spread over an area the width of the continental U.S., but we’ll stick to computing problems). The first problem is the data that this Square Kilometre Array (SKA) will generate. IBM estimates it will produce:

… a few Exabytes of data per day for a single beam per one square kilometer. After processing this data the expectation is that per year between 300 and 1,500 Petabytes of data need to be stored. In comparison, the approximately 15 Petabytes produced by the Large Hadron Collider at CERN per year of operation is approximately 10 to 100 times less than the envisioned capacity of SKA.

And guys, the LHC is in the midst of getting its own cloud computing infrastructure in order to handle its data. So this IBM/ASTRON project may be just the beginning for SKA. As I say in the headline, in many ways, projects like the LHC and the SKA are ambitious investigations into the origins and composition of the universe. Our investigations into dark matter will require a compute effort that could rival the engineering effort that it took to get men on the moon. Which makes big data our Sputnik and our Apollo 11.

Now, back to the problems associated with the telescope. It will generate data like a corpse breeds maggots, so the project needs a computer big enough to process it without requiring a power plant or two. Additionally that data might have to travel from the antenna arrays to the computer, which means the third problem is the network. I’ve covered the need for compute and networks to handle our scientific data before in a story on Johns Hopkins’ new 100 gigabit on-campus network, but the scale of the DOME project dwarfs anything Johns Hopkins is currently working on. From that story:

[Dr. Alex Szalay of Johns Hopkins] ascribes this massive amount of data to the emergence of cheap compute, better imaging and more information, and calls it a new way of doing science. “In every area of science we are generating a petabyte of data, and unless we have the equivalent of the 21st-century microscope, with faster networks and the corresponding computing, we are stuck,” Szalay said.

In his mind, the new way of using massive processing power to filter through petabytes of data is an entirely new type of computing which will lead to new advances in astronomy and physics, much like the microscope’s creation in the 17th century led to advances in biology and chemistry.

So we need the computing and networking equivalent of a microscope to enable us to deal with a telescope planned for 2024, and the time to start building it is now. That gives us a lot longer than the time frame we had to land on the moon. IBM views the problem as one worthy of the following infographic:

As the infographic shows, we’re going to need massively multicore, low-power computers, better interconnection using photonics and new ways of building our networks. Hopefully, the search for dark matter is worth it.

SKA image courtesy of the Square Kilometer Array.

Related research and analysis from GigaOM Pro:
Subscriber content. Sign up for a free trial.

Question Everything: A New Processor For Big Data

Stacey Higginbotham — Sun, 15 May 2011 21:30:55 +0000

We are moving from the Information Age to the Insight Age, as Parthasarathy Ranganathan, an HP Labs distinguished technologist told me. As part of that shift we need a computing architecture that will handle the storage of data, and the heavy processing power required to analyze that data, and we need to do it all without requiring a power plant hooked up to every data center.

The shift is a move from creating scads of information in a format that can be stored cheaply, to being able to process and analyze that information more cheaply as well (all the while adding new layers of data thanks to a proliferation of devices and networks). The challenge is that under the current computing paradigm, adding more processing is problematic both because it’s becoming more difficult to cram more transistors onto a chip, and those chips and their surrounding servers are sucking up an increasing amount of power.

With Power Consumption, The Question is, “How Low Can You Go?”

“Data is expanding faster than Moore’s Law and that’s a compelling problem that we’re trying to solve,” Ranganathan said. It’s apparently a problem that Intel’s Kirk Skaugen, vice president and general manager of the chipmaker’s Data Center Group, is thinking about too. Skaugen said at a speech last week at Interop that there were 150 exabytes of traffic on the Internet in 2009, and 245 exabytes in 2010, and the Internet could hit 1,000 exabytes of traffic by 2015 thanks to more than one billion people joining the web.

Want to hook one of these to your data center?

That’s a lot of bandwidth. But it’s also a lot of data and a lot of compute demand. Listening to Skaugen’s speech it appears that Intel’s primary function will be to convince the people who build the machines that process those exabytes of data, that their machines should run newer and more energy-efficient Intel processors. But is Intel’s architecture — and an upsell to its trigate 3-D transistors — the right chip for computing and big data’s future?

As I noted before, Intel’s much vaunted 3-D transistor advancement is cool, but only gets us so far in cramming more transistors on a chip and reducing the energy level needed. For example, a 22 nanometer chip using the 3-D transistor structure consumes about 50 percent less energy than the current generation Intel chip, but less than an Intel chip using the older architecture would at 22 nanometers (squeezing in more transistors also helps reduce the power consumption). And when we’re talking about adding a billion more people to the web, or transitioning to the next generation of supercomputing, a 50 percent reduction in energy consumption on the CPU is only going to get us so far.

The original, "flat" transistor at 32nm.

For example, scientists at the Department of Defense estimate (GigaOM Pro sub. req’d) that getting to the next generation of supercomputer at the current architectures would require possibly two power plants to serve every exascale computer — reducing that to one is great, but it’s not good enough. This is why the folks at ARM think they have an opportunity and why the use of GPUs in high performance computing is on the rise.

A New Architecture for a New Era

But there is more to this trend than merely eking more performance for less power — there is also a more subtle shift to matching your processors to your workloads in an acknowledgment that generic CPUs running x86 processors might not be the best solution for all workloads, especially in a cloud world. For example, startups are already trying to build optimized gear for companies such as Facebook or Google that can then run their own software on top of these optimized platforms.

Facebook's vanity-free server.

Don’t believe this is coming? Take a look at Facebook’s Open Compute efforts. This kept the same x86 processors made by Intel and AMD, but it was willing to question everything about the architecture of the servers and data centers those chips were house in. And that willingness to question everything is occurring at firms all over the world that are dealing with massive compute needs– a trend Intel can’t help but find worrying and folks such as Ranganathan at HP see as their big chance.

“Historically there is evidence that each killer app has an influence on the architectures that are preceded by the special purpose alternatives,” Ranganathan said. “So asking what instruction set for the processor, or if you want powerful or wimpy processors or special purpose processors are all legitimate architectural questions that we need to answer.”

HP’s answer is its concept of nanostores, chips that tie the memory and the processor together using a completely new kind of circuit called a memristor. The basic premise for HP is that 80 percent of the energy inside a data center is tied to moving data from memory to the processor and then back again. We’re already seeing the trend of moving memory closer to the processor (that’s what the addition of Flash inside the data center is about) to speed up computing.

But instead of next-door neighbors inside a box, HP essentially wants processing and memory married and in the same bed. HP won’t give a timeline on when this vision will become reality, but it has a manufacturing partnership with Hyinx it announced in 2010 to build such chips.

So Where’s Intel in This Architecture?

So when Skaugen gets up at Interop to push Intel’s 3-D transistors and the incredible inflows of data coming online he’s also making a pitch for Intel’s relevancy because big data processing is one of the areas where a general purpose CPU makes a lot of sense. So while folks may adopt GPUs for better supercomputing or data visualizations, or ARM may keep its upward momentum into more and more mobile computers or win some server designs in webscale businesses that can see a use case, just crunching those numbers associated with big data could become Intel’s game to lose.

There are plenty of folks hoping Intel will lose it (or at least that they will stand to gain) — not just Ranganathan at HP, but also the guys building 100-core chips at Tilera, or those hoping that the mathematical affinities inside digital signal processors might make them a good choice for data. It’s a topic I can’t wait to explore with Ranganathan, folks from Intel, Tilera and others at our Structure 2011 event on June 22 and 23. Because just like the steam engines and trains of the Industrial Age had to give way to the tools of the Information Age, the PCs and current servers used today will become a footnote as we pass into the Insight Age.

Related research and analysis from GigaOM Pro:
Subscriber content. Sign up for a free trial.

Adapteva Pitches A Supercomputer For Your Phone

Stacey Higginbotham — Tue, 03 May 2011 04:01:48 +0000

The brains inside your smartphone are getting more power with the latest application processors having two processing cores to help speed up the delivery of web site load times and mobile gameplay. That’s pretty awesome, but Adapteva, a Lexington, Mass.-based startup wants to take that number higher — a lot higher. The startup has created a design (and also the actual chip) for a 64-core accelerator that will sit inside a tablet or smartphone to help offload work from the application processor or graphics engine and do more computing on the device as opposed to sending it over a cellular or Wi-Fi network.

The concept of an accelerator is a familiar one in super computing, where the addition of a specialized massively multicore chip, such as a graphics processor or custom chip, is becoming more common. But unlike a GPU, the 64-core Adapteva chip only operates at one watt. To understand how powerful that is from an energy efficiency perspective, a four-to-eight-core server chip could operate at anywhere from 60-120 watts. And the challenge of building the next generation of supercomputers is constrained by the power demands such massive supercomputers would require.

But Andreas Olofsson, the founder and CEO of Adapteva, isn’t focusing on the HPC market at first–despite asserting that his design can scale to a 4,096-core design that would run at 64 watts. He said that while there is plenty of talk about low-power computing, “As long as you can plug something into a wall, the need for low power goes down significantly. It’s only a little bit painful.” However, in the mobile world where devices need to run all day, yet avoid bulky batteries, power consumption is at a premium. So that’s where Adapteva will focus for its big push (the company has some military applications as well).

The company began in 2008 and has managed to raise $2 million in funding from angels and boardmaker BittWare, its first customer. Amazingly, with that small amount of funding it has managed to have three versions of its chip built, making the startup incredibly capital efficient. However, the goal isn’t to build chips for the mobile market, but to license the technology, much like ARM, the firm behind the most common architecture in mobile phones, does. BittWare will manufacture the Adaptevea chip design — called the Epiphany– on its boards.

But in a highly competitive market, and especially on smarthphones, where space on the board is at such a premium, will device makers really embrace an unproven and as-yet-unneeded chip? Olofsson has two more difficult tasks to accomplish (since he’s apparently taken care of the hard task of building and designed a 64-core chip that runs at 1 watt for less than $2 million.) He must explain to board makers, chip firms and device makers why gadgets need this rather foreign accelerator chip, and he has to convince them that it makes sense to process data on the phone, rather than ship it over the cellular network.

The first task is made easier by the low-power envelope and by the fact that the full 64-core system on a system is fairly small — about 8mm square Olofsson said. Check out the model of the A5 system on a chip used inside Apple devices provided below to see how much space the Epiphany chip can take up. It would have to replace existing GPUs in this case.

The second task may be made easier by people’s desire to handle tasks such as speech or facial recognition or intense video games on their mobile devices. Olofsson argues that the latency inherent in sending even voice recognition to a server is problematic and that gameplay is impossible. Plus it costs more in terms of data charges and can drain the battery. “If you can keep the radio quiet and use the processing locally the battery life gets better,” he said.

Like many visionaries pushing a new technology he’s not entirely sure how the Epiphany could change mobile computing, but he’s certain that by boosting performance on smartphones to this degree it will. I’m eager to see if mobile device makers agree.

Related research and analysis from GigaOM Pro:
Subscriber content. Sign up for a free trial.

The Feds Want Faster, Greener Supercomputers

Stacey Higginbotham — Fri, 18 Feb 2011 15:13:21 +0000

President Obama’s budget is asking for $126 million for the Department of Energy to reach a supercomputing milestone — exascale performance. While supercomputing is an obscure branch of computing using odd benchmarks that even supercomputer experts sometimes debate, the need for supercomputers or high performance computing is only growing as we’re asking our machines to analyze more data, and appear on popular gameshows.

Supercomputers breached the petascale barrier back in 2008 with IBM’s Roadrunner and Cray’s Jaguar performing more than a million billion calculations per second, and thoughts immediately turned to the next obvious milestone, achieving one billion billion calculations a second. But instead of speeding up processors and the networking technology inside the machines, researchers are going to have to think first and foremost about power. Running an exascale supercomputer could require up to two city-sized power plants if scientist build the supercomputer out like the current machines. That’s not going to fly.

And because supercomputers now tend to run more mainstream components and software, the answers researchers find for their power and performance problems may be of use sooner rather than later for webscale computing or corporate IT departments. Perhaps we’ll see an ARM-based supercomputer or an entirely new architecture emerge if this funding is actually allocated. Maybe we’ll get a more power efficient optical networking technology, that could scale more cheaply for use inside the data center (going all optical saves power because it eliminates the need to switch a fiber optical signal back to an electrical one). The President’s budget may not pass Congress, but these dollars aren’t just about an obscure search for high-end machines, they could benefit and get benefits from more traditional computing innovations.

Related content from GigaOM Pro (subscription req’d):

Structure 2010: The Quest for Exascale Computing Power

Mathew Ingram — Wed, 23 Jun 2010 22:35:52 +0000

Supercomputer experts, including the chief information officer of NASA’s Ames Research Center and a computer strategist for the U.S. Army’s research and development center, said at the GigaOM Network’s Structure conference today that scientists are still working towards developing an “exascale” computer — that is, one that can do a million trillion calculations per second — to try and keep up with the flood of data that the world is producing every day, which continues to increase at exponential rates.

Chris Kemp, the chief information officer at Ames, said that the space program is also facing a storage issue that he called “the exabyte problem” when it comes to storing and processing the massive amounts of data it produces every day. “From Mars, we have Google Earth-resolution images coming down, and there are telescopes that generate an exabyte of data a day,” he said. “It gets to the point where you are forced to throw away data because you literally can’t store it all.”

But it’s not just NASA: Jason Hoffman, the chief technology officer of Joyent, noted that Apple recently announced it has sold 3 million iPads, “which means that they have basically shipped 100 petabytes of distributed storage in a matter of months.” When it comes to developing computers that can handle the processing of those kinds of data, however, science is running up against energy and cost issues, said John West, special assistant for computation strategy with the U.S. Army Engineer Research and Development Center.

The No. 1 supercomputing system today, he said, is the “Jaguar” system being used by the Oakridge National Laboratory of the Department of Energy to do climate modeling and other research. It can do 1.8 petaflops in terms of calculation ability, but it also sucks up about 7 megawatts of power to run, he said, and “these are $500-$100 million computers.” The Army scientist said that if you tried to get to exascale-level computing just by adding more processors, “you would be looking at 4 gigawatts of power and 125 million cores. We really can’t do this today.”

Chris Kemp also noted that having this much computing power has other impacts on storage and memory and other parts of a computer as well. “When you’re talking about that number of processors writing to memory, or writing to cache or writing to disk, it’s like the difference between driving across the street to Starbucks vs. going to the moon or going to Pluto in terms of the time it takes.”

http://cdn.livestream.com/grid/LSPlayer.swf?channel=gigaomtv&clip=pla_e91cb54d-d725-4a09-9cbb-7fa4a7cbc51a&autoPlay=false&mute=false

Watch live streaming video from gigaomtv at livestream.com